1. Technical Field
The present disclosure relates to an image tagging device utilizing a network represented by the Internet.
2. Description of the Related Art
In recent years, general object recognition based on a learning technique using a neural network called Deep Learning has been exhibiting high performance and attracting attention.
However, in order to achieve high recognition performance in general object recognition based on Deep Learning, it is necessary to perform learning processing of a large number of images, as the learning subjects, with which the name, kind, or the like of an object to be recognized is associated as a tag. Note that it is desirable that the tagged images be images in which the object indicated by the tag is located at the center or occupies a large portion.
In general, it is necessary to prepare approximately ten thousand tagged images, as the learning subjects, to achieve high recognition performance. This requires a lot of manpower and a large number of steps. Furthermore, since the number of steps is enormous, it is assumed that crowdsourcing is used for an operation of finding a portion necessary for recognition from an image such as a photograph or a video frame, cutting the portion out from the image, and tagging the portion. In this case, protection of privacy of persons in the images to be tagged is most important.
Japanese Unexamined Patent Application Publication No. 2007-213181 discloses a technique for generating a privacy-protected image whose image region for each user can be changed depending on the private policy of the user and that properly satisfies requirements for protection of privacy of an individual.